There were some questions a while ago as to if inverting the polarity of a signal was audible. After some discussionon head-fi, gaboo successfully ABX'd a polarity inversion. So far I'm completely unable to ABX his sample, but after poking around with another sample I was able to ABX a polarity inversion 10/10 in foobar.

Setup was as follows:
- Sound system was Chaintech AV-710 in high sampling 2-ch mode wired to Etymotics ER-4S phones
- Source material was 0:06-0:13 of Primus's "Hamburger Train" from "Pork Soda" (WAV ripped in iTunes)
- Inversion was performed with foo_invert, a plugin of my own devising. PM me for source if you want it. I verified its correct operation by doing a "quick mix" of the original and inverted WAVs in Audacity and verified the resultant file had all-zero samples.
- ABX testing was performed with only Volume Control in the plugin chain, 32 bit output, no dithering, 44.1khz.
- I did the ABX on the original 8:11 WAVs but only uploaded the 0:06-0:13 clip, so the filenames don't match the ABX result text. Hope that doesn't ruffle any feathers.

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/25 13:35:51

File A: file://C:\m\13 Hamburger Train.wav
File B: file://C:\m\13 Hamburger Train b.wav

13:36:17 : Test started.
13:38:44 : 01/01 50.0%
13:46:02 : 02/02 25.0%
13:46:57 : 03/03 12.5%
13:47:16 : 04/04 6.3%
13:48:11 : 05/05 3.1%
13:48:27 : 06/06 1.6%
13:48:42 : 07/07 0.8%
13:49:29 : 08/08 0.4%
13:49:46 : 09/09 0.2%
13:50:35 : 10/10 0.1%
13:50:39 : Test finished.

----------
Total: 10/10 (0.1%)

EDIT: What's the actual difference, you ask? It's best described as a very faint change in timbre in high-intensity percussives. It sort of sounds like a change in pitch, but I've found that's not really the best way to go about listening for it. Others have commented that low-frequency drum kicks seem to be sounding like they're coming from the opposite direction, but Etys are definitely not the phones to be listening for that sort of thing. This definitely was at the very cusp of my listening ability, but this 10/10 was on the very first time I tried testing with this clip.

Samples and plugin here.

Man, no offence, that hamburger hill stuff is horrible, can't get myself to listen to it even a few of times, let alone do an ABX.

Aw, c'mon! Face it, if you're already ABXing this sort of thing, you already like pain I probably ought to have done fade in/out on those clips too..

I tried ABXing again last night on my home system (RME PAD, Gilmore v2 amp, and either ATH-A900s or Etys) and I failed miserably through 16 or so trials. Then I did it again today from work in a couple minutes (AV-710 system) and got 8/8 on the first attempt. Go figure. So so far I have never missed a blind test with hamburger_train on my work system, but have always failed all tests on my home system. I'm starting to wonder if either there's something wrong with my work system, some sort of distortion that makes the reversal more audible, or something wrong with my home system that masks it. Or, my hearing may just be significantly degraded between the early afternoon and the late evening to the point that I can't hear it. I'll definitely need to look into this some more.

There are physiological reasons that can support audibility of inverted polarity, but there are restrictions to the scope.
It has been found that ear is sensitive to positive pressure, negative pressure being ignored. This leads to expectation that if signal is asymmetric wrt zero, it can become audible when inverting polarity.
If the signal is symmetric, no audibility is justified. Some instruments are known to produce asymmetric signal, so it depends on content.

I'm well aware of the physiological reasons for audibility, especially the rectification effect of the cilia. I don't think the signal needs to be asymmetric - the Wood effect was demonstrated by just tweaking the duty cycle of a sine wave, so that the last several degrees were zeroed out - which didn't affect the amplitude symmetry at all. But if the rising wave of the transient is of a much different amplitude than the falling wave then I could definitely see that as being a good candidate for audibility. And of course very loud bass kicks have periods so large that they ought to be good candidates too.

Moreover, my understanding of the literature is that so far it has only been experimentally detected under very strict listening conditions, very rarely with real CDs (much less Primus and consumer gear), and never on HA. Past discussions on the subject have not yielded a concrete conclusion, since nobody had really successfully ABX'd it.

(not that I'm quite claiming victory just yet. Once again I failed to ABX it at home, 4/12. Granted, I had alcohol in my system, and it was some ungodly hour of the night, so those are very extenuating circumstances, but still...)

Mods, is this the best forum for the thread? I was expecting a lot more discussion on this, given the 16/24 discussion...

OK, after a successful test tonight, I can more or less declare this as a fact: I can ABX hamburger_train, on the AV710, with the Etys. I cannot ABX it with the Etys and my RME/Gilmore combo. I cannot ABX it with my A900s and the RME/Gilmore combo. I haven't tried A900s and the AV-710 yet.

Tested to a multiple of 8 on the AV710, tested just now, in my usually fatigued state in the middle of the night:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/28 01:54:04

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

01:54:04 : Test started.
01:54:44 : 01/01 50.0%
01:54:51 : 02/02 25.0%
01:55:09 : 03/03 12.5%
01:55:16 : 04/04 6.3%
01:55:39 : 05/05 3.1%
01:56:04 : 05/06 10.9%
01:56:18 : 06/07 6.3%
01:56:24 : 06/08 14.5%
01:56:31 : 07/09 9.0%
01:56:37 : 07/10 17.2%
01:56:47 : 07/11 27.4%
01:57:03 : 08/12 19.4%
01:57:49 : 09/13 13.3%
01:58:25 : 10/14 9.0%
01:59:11 : 11/15 5.9%
01:59:25 : 12/16 3.8%
01:59:35 : 13/17 2.5%
01:59:41 : 14/18 1.5%
01:59:48 : 15/19 1.0%
02:00:03 : 16/20 0.6%
02:00:15 : 17/21 0.4%
02:00:21 : 18/22 0.2%
02:00:28 : 19/23 0.1%
02:00:37 : 19/24 0.3%
02:00:45 : Test finished.

----------
Total: 19/24 (0.3%)

And, as usual, utter failure on the RME, something like 4/8 right before the above test. It's not like I can tell any obvious distortion with the RME, like it's obviously messing up or anything like that - it's like the difference is just not there. This may not be an indictment of the RME, as it's quite conceivable that the AV710 is distorting in such a way that the reversal can become audible when it really shouldn't be. If so, then this really doesn't strengthen anybody's argument for audibility.

So, active hypotheses which I'll need to work on:
- RME is distorting and AV710 isn't, leading to masked polarity effect. To test: Find another ABX test which, if positive, virtually certainly indicates a difference due to the signal itself and not distortion; pass the test with the AV710 and fail it with the RME. Start with high frequency equalization testing. Alternatively, quantify any distortion present on the RME from the AV710 and write a plugin that replicates it; verify that the processed wavs cannot be ABX'd.
- Gilmore is distorting and AV710 isn't. To test: hook up Gilmore to AV710 and redo.
- AV710 is distorting and RME isn't, leading to nonlinear distortion with inverted polarity. To test: If the ABX fails with the AV710 running amped (with the Gilmore) then the issue gets rearranged into "is the Gilmore distorting?" and goto above.
- Both RME and AV710 are distorting but in different ways. This isn't really testable with current equipment.
- I'm trying harder on AV710 testing than on RME testing. Believe me, I'm not.
- Processing errors. I'm confident these did not occur and others can verify this with the uploaded flacs.

BTW, anybody's help with ABXing these or gaboo's samples are greatly appreciated.

I tested tonight with the AV-710, the Gilmore and the ER-4S. Test started off to 16, but due to my inordinate failure to get a positive result I decided to extend the test and try harder. I was convinced I wasn't just waiting for a positive result to come along due to chance at 25/34.

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/29 01:05:41

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

01:05:41 : Test started.
01:06:10 : 01/01 50.0%
01:06:17 : 02/02 25.0%
01:06:27 : 03/03 12.5%
01:06:49 : 03/04 31.3%
01:07:21 : 04/05 18.8%
01:07:28 : 05/06 10.9%
01:07:35 : 05/07 22.7%
01:07:53 : 06/08 14.5%
01:08:04 : 06/09 25.4%
01:08:39 : 07/10 17.2%
01:08:56 : 07/11 27.4%
01:10:57 : 08/12 19.4%
01:11:05 : 09/13 13.3%
01:11:17 : 10/14 9.0%
01:11:29 : 10/15 15.1%
01:12:30 : 11/16 10.5%
01:12:39 : 12/17 7.2%
01:12:58 : 12/18 11.9%
01:13:12 : 13/19 8.4%
01:13:39 : 14/20 5.8%
01:13:47 : 15/21 3.9%
01:14:04 : 16/22 2.6%
01:14:19 : 17/23 1.7%
01:14:57 : 18/24 1.1%
01:15:14 : 19/25 0.7%
01:15:23 : 19/26 1.4%
01:15:31 : 20/27 1.0%
01:15:52 : 21/28 0.6%
01:15:59 : 22/29 0.4%
01:16:07 : 23/30 0.3%
01:16:25 : 23/31 0.5%
01:16:38 : 23/32 1.0%
01:16:57 : 24/33 0.7%
01:17:36 : 25/34 0.5%
01:17:57 : Test finished.

----------
Total: 25/34 (0.5%)

So I am able to ABX with the amp; the odd man out here is the RME. I assert this immediately eliminates the Gilmore or the output stage of the 710 as being likely sources of distortion. (One thing I failed to mention earlier is that I was using the 710 in high-sampling-rate mode, on the rear speaker outputs, but I was NOT resampling to 96khz during the ABX.) This leaves either the RME DAC or the 710 DAC.

As a quick test to make sure the test wasn't corrupted by playing the tracks at 44.1 instead of 96 as suggested by the drivers, I did a quick test at 96 using SSRC:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/29 01:33:41

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

01:33:43 : Test started.
01:34:55 : 01/01 50.0%
01:35:13 : 02/02 25.0%
01:35:18 : 03/03 12.5%
01:35:25 : 04/04 6.3%
01:35:46 : 04/05 18.8%
01:35:52 : 05/06 10.9%
01:36:02 : 06/07 6.3%
01:36:07 : 07/08 3.5%
01:36:12 : Test finished.

----------
Total: 7/8 (3.5%)

To determine which of the two could be at fault, I propose one "easy" test and one "hard" tests. The first "easy" test is to implement a polarity test that is known to pass on any reasonable hardware, using artificial signals. I'll first try the Wood effect (clipped sine waves) and pulses of opposite polarity. The "hard" test is to locate a suitably precise data acquisition card, put both cards under the knife, and try and numerically identify a distortion effect which could result in the observed results.

Advice: You're going to need to produce better than 7/8 results.


Note: Not criticism, just advice

Just a note on ABX testing: the number of total trials should be decided on before starting the test. Initially deciding on 16, and then monitoring the results until you get a desirable endpoint is a form of "cherry picking."

If you're interested in ABXing a small effect, and don't want to err by saying "there's no difference" when in fact there is one, you need to be prepared to perform a lot of trials.

This beta version of abchr has a training abx mode, will hide the in-progress results for a real test, and calculates required trials to get desired error risks given the size of the effect.

http://ff123.net/export/abchr1.1beta2.zip

ff123

OK, from now on I'll fix the number of trials at 32 for all tests. I'm fairly confident I can ABX consistently at that N.

Basically ABX is a scientific tool/statistical tool which enables you to prove that there is a statistical difference between two different audio files. This requires quite a few attempts and successes to be statistically sound. The assumption is made before the test that the two files will not produce an audible difference but when a certain confidence level is reached either .1 or .01 as a result of a number of successful trials, only then can two files be scientifically/statistically proven different in their sound.

GUys, I don't need a lecture on blind testing, I know exactly what's involved. First of all, in the context of most listening tests conducted on this site, my understanding is that a confidence level of 0.05 is considered acceptable, and I'm clocking in waaaaay under that - <0.005, to be exact. And I do understand how cherry-picking the stopping point for the test can reduce the significance of the results.

But how can my results be weakened beyond significance if the significance I'm already at is very strong? ie, it's idiotic to throw out a positive test where the stopping point was cherry-picked at N>100 and p<0.001 - at what point does the stopping point matter? Moreover, if I was cherry picking the place to stop, wouldn't some tests necessarily need to be conducted an indefinite number of times due to chance?

My only confusion appears to be in the nature of defining the number of trials, and even then I don't think my current results are significantly weakened. I will admit to a very slight bit of cherry picking on the 25/34: I missed one trial, exited, checked my config (and didn't wind up changing anything). IIRC every other test was from the first trial on.

In any case: this is no longer an active issue, I'll make sure the tests I do in the future are N=32.

OK, I've had time to do N=32 tests on both cards, and finally, I can kinda-sorta claim positive ABX results on the RME. Configuration: volume levels set to whatever I felt was appropriate to do testing and were allowed to be changed mid-test. Card was routed through RS Gold 1/8-RCA stereo cable to Gilmore V2, then on to ER-4S phones. RME was set to -6dB and AV-710 was set to full volume. ABX was done through foobar with SSRC resampler at 96khz and no other DSP components. Absolutely no tests were dropped and no training was done before the tests.

The AV-710 test was done first:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/31 19:12:14

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

19:12:16 : Test started.
19:13:18 : 00/01 100.0%
19:13:55 : 01/02 75.0%
19:14:20 : 02/03 50.0%
19:14:29 : 02/04 68.8%
19:14:48 : 03/05 50.0%
19:14:56 : 04/06 34.4%
19:16:31 : 05/07 22.7%
19:17:05 : 06/08 14.5%
19:17:29 : 07/09 9.0%
19:17:40 : 08/10 5.5%
19:18:15 : 09/11 3.3%
19:18:31 : 10/12 1.9%
19:18:48 : 10/13 4.6%
19:20:40 : 11/14 2.9%
19:21:03 : 12/15 1.8%
19:21:19 : 13/16 1.1%
19:21:32 : 14/17 0.6%
19:23:30 : 15/18 0.4%
19:23:42 : 16/19 0.2%
19:24:07 : 16/20 0.6%
19:24:31 : 17/21 0.4%
19:24:39 : 18/22 0.2%
19:25:09 : 19/23 0.1%
19:25:17 : 20/24 0.1%
19:25:40 : 21/25 0.0%
19:25:47 : 22/26 0.0%
19:26:15 : 23/27 0.0%
19:26:23 : 24/28 0.0%
19:26:31 : 25/29 0.0%
19:26:39 : 26/30 0.0%
19:26:48 : 26/31 0.0%
19:27:07 : 26/32 0.0%
19:27:14 : Test finished.

----------
Total: 26/32 (0.0%)

No surprise here. Next the RME:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/31 20:03:04

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

20:03:05 : Test started.
20:04:47 : 01/01 50.0%
20:05:06 : 01/02 75.0%
21:06:53 : 02/03 50.0%
21:07:00 : 03/04 31.3%
21:07:27 : 03/05 50.0%
21:08:19 : 04/06 34.4%
21:08:27 : 05/07 22.7%
21:08:38 : 05/08 36.3%
21:08:58 : 06/09 25.4%
21:09:09 : 07/10 17.2%
21:09:32 : 07/11 27.4%
21:10:06 : 08/12 19.4%
21:10:17 : 08/13 29.1%
21:10:29 : 09/14 21.2%
21:10:59 : 10/15 15.1%
21:11:11 : 10/16 22.7%
21:11:42 : 11/17 16.6%
21:11:50 : 12/18 11.9%
21:11:57 : 13/19 8.4%
21:12:13 : 13/20 13.2%
21:12:40 : 14/21 9.5%
21:12:59 : 15/22 6.7%
21:13:11 : 15/23 10.5%
21:13:26 : 16/24 7.6%
21:13:34 : 17/25 5.4%
21:13:41 : 18/26 3.8%
21:13:53 : 19/27 2.6%
21:14:04 : 19/28 4.4%
21:14:15 : 20/29 3.1%
21:14:27 : 21/30 2.1%
21:14:42 : 22/31 1.5%
21:14:54 : 23/32 1.0%
21:14:56 : Test finished.

----------
Total: 23/32 (1.0%)

Finally, a halfway-positive ABX result on this card (p<0.01). The primary artifact I've been listening for on the Chaintech has been a raise in timbre/pitch on the inverted sample on the drump strike at the theme transition, about 3.5 seconds in. This is next to impossible to hear on the RME. Eventually I figured out another effect, a sort of increase in air/loss of dynamic range between the drum hits, which contributed one or two points to the final score when I couldn't figure out X/Y from the drum strikes alone.

I can't help but agree with the subjective opinions at Head-Fi that there is a general muddiness in the high end on this card, because overall the sound felt a bit more "toppy" than the AV-710, and not in a good way. Nevertheless the difference between the Chaintech and the RME is very, very hard to spot, and I don't think I could notice it in a nonblind test.

Nevertheless, the 3 points less result on the RME belied the fact that I had a much harder time with the test than on the Chaintech. With the Chaintech, I could often reach a firm conclusion after listening to the sample for less than a second, and when I didn't, it was usually because I either got the last round wrong or I (rarely) missed the timbre change and needed some more time. With the RME, every test was a long struggle. I hope this is enough of a reassurance to assuage my own complaints about this sort of blind test, but again, you basically need to take me on faith that I tested both cards equally hard.

About the only thing left I'd like to do is make sure this is audible on my A900's. Then I'm claiming victory. I don't think it's necessary at this point to delve into objective testing between the sounds of the RME and the AV-710; everybody here seems to be OK with subjective descriptions between encoder samples on positive ABX results, so I think it's reasonable to give subjective descriptions on negative (or close to negative) ABX results.

Tests for Audio Technica ATH-A900, Chaintech AV-710, everything else same as last post.

foo_abx v1.2 report
foobar2000 v0.8.3
2004/10/31 21:56:40

File A: file://C:\hamburger train.flac
File B: file://C:\hamburger train inverted.flac

21:56:42 : Test started.
21:57:18 : 01/01 50.0%
21:57:25 : 01/02 75.0%
21:57:33 : 01/03 87.5%
21:58:14 : 02/04 68.8%
21:58:34 : 03/05 50.0%
21:58:45 : 04/06 34.4%
21:59:38 : 05/07 22.7%
21:59:54 : 06/08 14.5%
22:00:07 : 06/09 25.4%
22:01:10 : 07/10 17.2%
22:01:26 : 08/11 11.3%
22:02:23 : 09/12 7.3%
22:02:31 : 10/13 4.6%
22:02:44 : 10/14 9.0%
22:04:20 : 10/15 15.1%
22:04:52 : 11/16 10.5%
22:05:00 : 11/17 16.6%
22:05:17 : 12/18 11.9%
22:06:18 : 12/19 18.0%
22:06:35 : 13/20 13.2%
22:06:51 : 14/21 9.5%
22:06:59 : 15/22 6.7%
22:07:07 : 16/23 4.7%
22:07:40 : 17/24 3.2%
22:07:47 : 18/25 2.2%
22:07:59 : 19/26 1.4%
22:08:10 : 19/27 2.6%
22:08:30 : 19/28 4.4%
22:09:15 : 20/29 3.1%
22:09:47 : 20/30 4.9%
22:10:16 : 20/31 7.5%
22:10:24 : 21/32 5.5%
22:10:26 : Test finished.

----------
Total: 21/32 (5.5%)

Grumble. Important results, but not significant enough. The change in timbre was not nearly as noticable as even on the RME with the ER-4S, and I had to instead rely on more intangible quantities like "air".

What do you guys think? Are my results significant enough as they stand? What else would you like tested that would be reasonable for me to test?

I believe most power amplifier cascades actually have two parts -- one transistor amplifying the upper half-wave, another one amplifying the lower half -- one being p-n-p, the other one n-p-n. Because it's fairly difficult to manufacture two semiconductors on a different base with exactly the same amplification slopes, it's very possible to have lower half of your signal amplified slightly more than the upper half, or vice versa -- thus making even a perfectly symmetric signal asymmetric.

Even if you have just one regular transistor amplifying your signal (pre-amp cascade), the output can still be asymmetric if the bias current is not tuned perfectly -- and besides, even in the vicinity of the bias current, the amplification curve of a transistor is not exactly a straight line, nor is it perfectly symmetric about the bias current.

Moreover, the membranes and the attached coils suspended in the magnetic field, comprising your headphones or speakers, may also have greater elasticity/sensitivity with respect to bending outwards as opposed to inwards or vice versa, depending on how they are suspended -- just like your eardrums.

This would explain why your ABX is so much dependent on the equipment you are using -- with this sort of thing, you are essentially testing your equipment as much (if not more) as your hearing.

You are completely wrong. You would have first a basic knowledge in electronics before giving an opinion about topics that you don't know well.

Today the audio amplifiers don't suffer of this problem, because the negative feedback compensates all these nonlinearities; with this you obtain a perfect symmetric output signal; with levels of THD (total harmonic distortion) so low that they are only noticeable with measuring instruments.


The modern audio circuits are composed by many transistors, not only one, and the bias is tuned correctly; and, in the majority of the cases (in low level signals) this is done with operational amplifiers composed with a lot of transistors with correct bias. Even in the case of final power amplifiers stages (clase AB) the effect the crossover distortion is completely nonexistent, as a result of a good combination of adequate bias current and negative feedback; all these are basic factors of design taken in consideration in today electronics.


Wrong. Your arguments don't have sense at all. The cause is not the audio equipment.

I do have more than a basic knowledge in electronics, although my knowledge may be out of date, so save your breath; explaining where I was wrong here is sufficient.


It's all well and good in theory, but a lot of things happen that aren't supposed to so long as the possibility exists.


You still have not answered the headphone/speaker part. Care to exercise your basic knowledge in physics?

Why did you put the AV-710 in high sampling mode for a 44.1 kHz track?

Higher 'quality'. Just kidding. Perhaps there is a valid reason.

I know that the guides say to use HS mode to get output through channels 7&8, which are attached to the WM DAC, but the new drivers allow 7&8 output in regular 2 channel mode (see Wish's post here). Using 2 channel mode would rule out the possibility of some sort of resampling quirk causing the difference.

Axon: As only Volume Control was in the DSP chain I'm assuming you let kMixer handle the resampling?

If you guys look at the last time axon posted, it was october last year! While i'm sure he's still around, that would explain why his tests were done in "high rez" mode because like you said, the driver's that do 16/44.1 on the wolfson dac probably didnt exist then.

Last night i re-read through this thread. Then this mornining with a clear head, i did some ABX testing on my speakers.

My setup is e-mu 0404 > tripath amplifier > klipsch bookshelf speakers. I was easily able to hear what Axon described as a change in timbre.

I used the e-mu's phase invert which is just a button you press "on/off" and it changes the phase/polarity on the fly gaplessly. This is assuming e-mu's phase invert does the same thing as switching polairty 180 degrees.

Although I didnt ABX myself because i could see if the phase switch was on or off when i clicked it. It really really did sound different. Not in a placebo way. Since I can't prove it till I try out Axon's fb2k plugin, try and take my word for it too, the difference is there. If you have an e-mu card try it out for yourself.

-Joe

You guys are testing your equipment!

Damn straight! reversing polarity has much more darastic effects on headphones vs speakers.

2-driver loudspeakers tend to have crossovers where the tweeters are 90degrees off one axis and the woofer 90degrees the other way. So when reversing polarity, your really only reversing the polarity of your tweeter in conjunction with the woofer.

Headphones are a whole different deal, you can lump in headphones with full range drivers for loudspeakers because they use one driver to handle the whole "range". It might not be obvious with a straight up tone, but once you start introducing multiple tones you get "doppler distortion" and once this happens you notice polarity really quickly, that's why full range drivers tend to have lots of frequency modulation distortion.

Here's an example to help see what i'm talking about. If you have a kick drum and a flute duet. The kick drum moves the air forward while the flute rides on top of it. Now if you reverse the signal the driver moves backwards while playing the very same flue signal. But since sound's frequency increases as it moves twoard the listener and decreases as it goes away (think a train passing you) you basically end up changing the hanging pitch of the flute when you do this with full range drivers because the cone motion is so drastically different beween the kick drum and flute frequencies.

Speaker users will notice this, an interesting concept is that we don't just listen to sound with our ears...we feel it in our chest too. If you get the polarity backwards on a kick drum it starts to sound wierd because you're expecting a compression wave to hit you first, not a trough followed by a smaller compression. Put your hand in front of the speaker and feel the air move by when the kick hits...then reverse the polarity and notice how much it changes. You practically don't feel it when the woofer moves away from you at first.

Btw, it doesn't matter if absolute polarity was maintained throughout the entire signal chain (studio/mastering facility etc etc) because the relative polarity of everything in the mix will be the same throughout the chain. It's then up to the listener to determine when the polarity should be switched, but I can't think of any recordings that I own though that might require this.

If the equipment makes this audible, so what? That is irrelevant in the context of justifying phase correctness in mastering and playback. The only possible argument that could be made is whether the wrong polarity is necessarily any worse than the correct polarity, which doesn't really weaken the importance of the results any. Nobody in the literature has even been able to ABX this on real music. Sine waves, sure, but not Primus. In fact, if this were an equipment issue, that might make the results more important, depending on if different amplifiers result in different levels of audibility. (It would to date be the first known way to objectively measure audible distortion with modern, properly functioning amplifiers.)

Anyways, given the failure of most/all ABX tests against amplifiers, compared to the very well known and documented Wood Effect, I would much more easily believe that the reversal is audible than believe a problem with my system. I could be persuaded otherwise.


The Wolfson DACs on the rear outs are considered objectively superior to those on the front stereo channels.

Actually I think that my card could accept 44.1. And I'm 99% sure I avoided kMixer on this chain, because the rear output mode doesn't even accept mono signals as an input - you need to convert to stereo explicitly. If kMixer were there it should have been able to convert automatically.


Doppler distortion only applies to loudspeakers because it's supposed to be correlated to the velocity of the driver, and extremely long displacements are necessary to pull it off. So far, not only has it not been ABX'd AFAIK, I'm not even sure Stereophile believes it's audible.

I certainly agree the WM DAC is far superior. I'm playing 44.1 kHz music in normal 2 channel mode through the rear outs using the latest Via drivers and foobar2000 in KS mode. Electrically the Via DACs are not connected to the rear outs.


The card can accept 44.1, that's my point, but only in normal 2 channel mode. By setting it to high sample rate mode you are restricting it to >96 kHz. Try playing a 44.1 sound file in foobar2000 using KS in high resolution mode - it will give you a KS output error, because the card will only accept >96 kHz audio in high sample rate mode.

The reason you can play back 44.1 in high sample rate mode through DS or waveOut is kMixer is resampling the audio to 96 kHz. The card can not accept anything less in high sample rate mode through any output method.

Try putting it in normal 2 channel mode, then use KS in foobar2000 to feed it 44.1 kHz sound, it should play correctly through the rear output. If not, try grabbing the latest Via drivers, because it's working fine for me.

OK, now I'm just confused on exactly what I did and exactly what could be supported. I'm quite certain I used Kernel Streaming, but I'm not sure if I resampled, or if I fed 44.1 into KS. I'd guess I did the former, probably with foo_pphs.

If you used KS, then you should have gotten the below error unless you have PPHS or SSRC in your DSP chain. Anyway, try putting it into normal 2ch mode and removing any resamplers from the DSP chain. Probably won't make any difference to audiability, but if you can rule something out easily you may as well do so.

I'm well aware of what it would look like.

Oh well, maybe I'll do this again sometime, but I'm not holding my breath. It's not a high priority for me to reproduce this, especially since nobody else is giving it a shot (wink wink).

Axon, when we have equipment that makes something audible depending on phase, then we have faulty equipment. Not in a fatal sense, but in sense of not being completely transparent. Knowing that our equipment isn't perfect, how can we even consider arguing about audibility or correctness of phase?

Phase correctness in mastering and playback is a matter of standard. Its so trivially easy to maintain, that there is no justification whatsoever to phase reversal in the recording chain. I consider this part of the matter irrelevant. Incorrect phase on a recording is simply huge label of disgrace for the recording company. Its a sign of negligent attitude. What quality can you expect from a recording company that can't even get its phase (wires) right?

And the part of equipment that makes it audible is not amplifier, but electromechanical system, electromagnetic to acoustic transducer we call speaker. Headphones are less influenced, but still are too, especially those that have wild riding frequency and phase response.
Perhaps you heard cheap electrolytic capacitors with your 710, they can have asymmetric response.

The effect that bothers most is dependence of magnetic force on the cone as it is displaced from its rest position. Not only is the force lower to either side from the center, but it is also almost always *asymmetric*, which in quite many cases causes the cone to push out and stay offcenter, depending on the music material. Its there where phase reversal changes mechanical behaviour, and as being off the optimal position imparts distortion, the nature of distortion depends on where or how much the cone is displaced.

Have a glimpse of how bad it really is with speakers:
http://www.klippel.de/pubs/assesslargsign/...performance.pdf

Anyway, there is single strong argument against audibility of continuous phase changes - it happens all the time as you move in the space. If we were sensitive to phase, then we'd hear timbre of the world around us differently every time we moved our heads, every time we moved a footstep. That alone makes the race for perfect phase response a moot, imho.

Things we hear are really different distortions, not the phase reversal itself. Thus we can't really even know which of the two phase positions is closer to transparent - either could be wrong.

If that is your standard of perfection, then we must have a lot of defective hardware floating around. Much more than would be expected from the quantitative distortion measurements and ABX failures of various amplifiers.

Now, I do agree that your hypothesis is pretty likely, but at this point I can't discount the alternative hypothesis (that the effect is real and that some sources have other types of distortion that mask it). Do you have any suggestions on how to distinguish between the two cases? The only thing I can think of is to maybe put the card under the knife and measure its inverted/noninverted distortion with a good digitizer.


Actually, most audiophiles who care enough to look at the waveforms looking for proper phase have stated that most/all recordings in existence have wrong phase in at least one of their tracks. In other words, most music is recorded in both correct and incorrect phase.

A lot of highly innocuous things can mess up the polarity. Many (perhaps most) preamps invert phase, since inverting amplifier configurations as a rule of thumb have less noise than noninverting amps. Speakers are not necessarily wired for +/+ polarity.


Whoa there! You can't have it both ways. Either the transducers are the only source of the distortion, or the amplifier may possibly be the source of the distortion. But the two statements are logically contradictory.


Thanks, this is really good information. This is sort of corroborated by the fact that I need to listen at loud volumes to notice it. However, based on the rated specs of my headphones, I'd need to exceed 120dB SPL to exceed them, so the plausibility of this being an issue when my ears don't bleed after the test is not established.


Wrong. You are completely misunderstanding wave mechanics. You cannot invert phase by moving around in space, at least in the presence of a point or planar sound source.

I would explain why, but somebody on Head-Fi made the same mistake, and I and somebody else laid the smack down on him for several days and he never seemed to figure it out. So please ask if you want it.

The example is also quite flawed because I notice obvious changes in timbre as I move my head around, but that's due to HRTFs and pinna effects.

In serious? How could you do this?

Which possibility? For your information; my statements HERE are not "possibilities" like either "black magic". They have been verified by engineers by more than 30 years, and they are basic laws of the electronics.
You responded to my post with evasive words, with the intention to turn aside the attention about your lack of knowledge about the topic.


You are wrong again.
I didn't respond to this part of your post because I considered it irrelevant, not by another reason.
The good speakers and headphones are designed with the goal of to reproduce the music signal as faithfully as this is possible, and the only way to obtain this (for a mechanical device like a speaker) is that both movements (outwards and inwards) have similar elasticity/sensitivity behavior.

And why would you think otherwise? And again, why do you keep the amplifier as the only equipment of interest? There are not many transducers whose THD is even in same order of magnitude with cheapest and crappiest 1-bit DAC or fullrange amplifier..

First, we need to understand what we are looking for - what at all can make it possible to hear phase inversion, ie. turn to physiology of ear and hearing. Then we'd like to check whether limited recording chain is even capable of providing us with such signal, then we can think of method to detect fault in hardware, deal with it and evaluate hearing.

If you can point to specific place in your sample where you hear it, it would be nice to try to pinpoint it to very precise section of wav, extract it and analyze it. If it appears to be possible to reproduce with synthetic test signals, then we are onto something. We'd understand correlation to audibility. You could run that section through loopback test to try to detect measurable changes.

To try to capture the fault, I'd first try to make sure whole reproduction chain has same slew rate on both positive slope and negative slope. Depending on implementation it could go from perfect to awful. Price doesn't matter.

There are many faults that *could* be the cause, but of these many things *most* are inaudible. And hard to measure because they occur in specific dynamic conditions only, thus no formal generally accepted methodology exists.
Currently lots of debates are around transient distortions that are not measured with the THD methodology. They manifest only during fast changes of amplitude and spectrum, adding or delaying spectral components only for that transient duration. Perceived changes in pitch hint on phase modulation or pretty complex intermodulation distortion components near the fundamental you are listening to.

In terms of real audibility, I can only think of asymmetric nonlinearity inside our hearing organs that could *sometimes* trigger audibility of phase reversal. But it is so rare event that its completely pointless to chase.
If we could ABX phase reversal every single time we faced it, please, but if we have to seek for that incredibly rare example where its audible after an incredible effort and still with very arguable reliability, without slightest evidence its not due to hardware, then excuse me, we are chasing ghosts of pigs that maybe could have flied in their former life. This isn't empirical evidence, its ancedotal evidence.

Your own drastic difference in reliability to ABX your samples on differing hardware gives a straight hint that its about hardware, not ears. And you have ability to deductively find out which piece of equipment it is. Its not that some subaudiophile equipment is "masking" total absolute phase reversal, is it. How'd you mask something like that anyway?
And you believe this?

But this is fancy playback chain, not recording..
Not that your points on most preamps or rule of thumb is true..
Your logic sucks. For $25 you get what you pay for. Be happy that it has so increbily good sound for that money, but please don't take that for granted.
And 710 hardly qualifies as amplifier. I just offered you a possible extreme case of crappy component that certainly wouldn't be expected in $500+ amp but is *easily* acceptable in a pretty complex 8-channel device as 710 for $25. I mean, $3 per channel? Come on, you can barely get 8 cheapest RCA cables for that money.. and you expect to reliably evaluate physiology of hearing with that?

Of course amplifier *CAN* be audible source of distortions, I never said opposite. I said that most likely it isn't, that much higher distortion is produced with that last passive mechanical device. You can't be talking about real audibility until you can show that it is not the distortion of that last device you are hearing.

You focus on wrong thing. At loud volumes it becomes catastrophic - 30% THD and such. Look at the graphs and think of shapes - they are there at low volumes aswell, just smaller.

You'd want to observe how magnetic force is asymmetric, how it drops upon any kind of displacement off the rest position, how compliance is asymmetric. Finally notice how wildly the distortion depends on loudness, frequency - ie. dynamics. The paper doesn't even touch intermodulation distortion that motor creates depending on driving methods, etc. Its a mess. Headphones are quite different in any case - mechanics is very different. So you can't apply that paper directly to headphones. But they have their own list of problems.

But interesting where you got the 120db requirement for your phones? Do you have any data about displacement of membranes of your headphones? Do you have even remotely reliable data about their distortion I wonder?

I believe you are not in position to judge my "complete misunderstanding" of wave mechanics. I'd suggest you to first make some effort to understand what other side means. And don't say that because I didn't write an essay you are free to assume whatever you like.
You assumed that I assumed that absolute phase inversion occurs. You assumed wrong.

Ever heard of room acoustics? Have you found a point source? Have you ever looked for correlation between magnitude of phase changes and perceptible timbre changes? There are works on that matter, and only very drastic and abrupt phase response changes are found to be audible. The rest goes into audiophile myth area.

Can you tell how much does phase of 10kHz wave changes wrt to say 200hz when your distance to tweater changes 3cm while distance to woofer stays the same? And tell how obvious effect on timbre that has.

Evolution has arranged our hearing so that we are generally insensitive to absolute phase and to a large degree changes to phase of spectrum components so that we can recognize and distinguish sounds by timbre, and use phase of transient's envelope for localization. It would be distractive if timbre changed in proportion to phase changes. We wouldn't survive in the wild.

I hope you don't insist that phase response in your room doesn't change as you move? Total inversion is just very special case, not even relevant to timbre..

Splitting up my reply into several parts.


I apologize for my tone there. I do admit it was out of line.


Ow. I thought for a while that maybe we were confusing phase with polarity, but after doing more research I realize the source of my confusion: I've always thought that waves reflected from surfaces are of the same polarity. After reading this, and remembering what happens to untaut lengths of string when you thwap them on one side and watch the reflection happen on the other side, I realize I was mistaken: reflected waves are of opposite polarity.

So you can consider my foot very firmly in my mouth right now

If I assume that reflections are of the same polarity, then obviously you can't construct any situation with passive materials that would negate the polarity of a signal. Which completely explains our gap of opinion. Besides that I totally agree with you on the audibility of phase shifts and group delay. Short of pathological cases like standing waves, they are not that audible, if at all.

Now, hold on there.

Pure time delay, also known as linear phase, would be very confusing indeed if we somehow could hear it. Of course, we don't hear pure time delay, and we don't even hear something that looks somewhat like a pure time delay over a critical bandwidth or so on the basilar membrane.

If you said "to some degree" in your comment on phase, I'd be content. We just had a rather substantial argument somewhere here about phase audibility.

Narrowband FM vs. AM (both have exactly the same power spectrum) do not sound the same, and you can demonstrate this trivially with Octave or Matlab.

In that thread, there was also some dispute over the validity of Fourier Analysis, (sigh) but I don't think we have that problem in this thread.

The key, I think, is to learn to separate out the "delay" part from phase shift. When you talk about moving to and from a speaker, you're simply performing a linear-phase behavior, and that's not going to change envelope, waveshape, etc, except by pure time delay, ergo that discussion is just not germain.

A germain change in phase is the same as having substantial dispersion in the signal, i.e. different frequencies get there at different TIMES.

I'm not sure what you say here. I was NOT talking about "moving to and from a speaker", nor about pure time delay.
About your narrowband FM/AM I'd like to discuss some more in your thread. But it is example of "very drastic and abrupt phase response changes".

OK, you've got me stumped again. Nobody seems to have actually pointed out "X recording was made in wrong polarity", and it's unclear to me how to clearly distinguish between wrong polarity and moving the mic to opposite where it was before.

Positive Feedback mentions a record which was flipped in polarity by the mastering engineer, which implies that either that rereleased LP or the original LP was in the wrong polarity. But that's kind of a pathological case.

I'm going to ask around and see if anybody knows of some known CDs with bad polarity, and I might try and identify reversal with vocals across multiple CDs, although I'm unsure of the soundness of that approach.

Actually, I was referring to the variability of characteristics and the probability of manufacturing defects that exists any time mass production is involved. No one tests every single transistor or chip; only a sample is tested -- and neither is it feasible to test the entire assembled device in all possible operation modes -- only to perform some basic tests to verify that the assembly is OK and that no components are failing outright.

And even though the variance in characteristics and the probability of failure of any individual component is fairly low, not enough to be noticeable without measuring instruments or operating the component outside the spec, when combined to a single device and given the amount of components involved in a typical piece of electronics, the probability of the whole device not behaving by design approaches 1 feairly quickly. Anyone who ever dealt anywhere near closely with any piece of electronics, analog or digital, knows that rarely does it work completely as expected. Sorry, but that's just about the only thing that has been "verified" by me personally in the past 15 years, so your reassuring words miss my ears.

"Possibility exists" in this case referred to negative feedback, which is a workaround for the problem rather than a solution for it -- and besides, you can only push it so far. The more feedback, the faster you converge, but the more you drive your gain down at the same time. The more you drive gain down, the more cascades you need, and the more cascades you need, the more non-linear distortions (which you are trying to compensate for) you have in the first place, no matter whether you loop the entire system or just the individual cascades. Besides, the more cascades you have, the more noise and positive feedback you have circling around in the power bus, hence the more filtering you need. Power filters are (were?) usually just RC filters rather than LC, and tend to displace the bias, making it more difficult to tune with the usual 5-10% variability in resistances. So -- not sure about now, but 10 - 20 years ago, finding the right amount of negative feedback (and the right amount of cascades) was a compromise between quality and performance. And since the basic principles of operation remain the same to this day, I doubt anything has drastically changed here -- maybe now you can use stabilizers in place of RC filters and operational amplifiers much more widely because you can put so much stuff on a chip.

One other thing we have not mentioned so far as "by design" behavior is concerned -- a number of factors that have a profound effect on the operation of semiconductors (especially in analog circuits) that are not completely under the designer's control, the most prominent ones being temperature and age -- which, even though they can be _somewhat_ compensated for with additional circuitry or toleranced, cannot truly be controlled for.


Intentions is an ambiguous matter because you can read between the lines in any text, and "find" meanings the original author never even thought about, but which in reality are your own wishful thinking. Intentions are therefore irrelevant for the purpose of this discussion; actions (and their immediate causes) are what matters. So let us cut the guesswork about each other's intentions and remain on topic please.


That's right -- but in a perfect speaker, the magnet, the membrane, and the coil must be perfectly symmetric, and the area of the diffusor itself must be infinite -- which of course is not the case. Again designed with the goal, but in practice only approaching the goal. And just like with the transistor amplifiers, you can slap on some fixes to mask the problem, but not really fix it. Such, one might design a particular shape that would behave "reasonably" well in reproducing a wide range of frequencies "approximately" independent of the source and destination position; put a box around the outer edge of the membrane to prevent lower frequencies cancelling themselves, pick the shape of the box such that the resonances compensate for the speaker's otherwise lousy frequency response, etc, etc -- all that is being done, I don't doubt that. But there are simply too many factors involved (many of them outside the designer's control) to create anywhere near complete mathematical model and find an anywhere near perfect solution.

The explanations that others have put forth all (most?) sound reasonable, but I guess we'll never know whether it's hearing or equipment specifics that allow one to hear polarity inversion -- I still tend to think one as much as the other, especially that the majority of the reasons seem to be just plain physics rather than psychoacoustics. Probably the only thing made clear is that people have their DACs set to top notch, so digital->analog conversion is probably not the cause.

For my Etys, the max rated SPL is 122dB. I guess I pulled 120db out of my butt for the A900s (I do remember it being around 120) but I might be able to dig up the specs on that.

I would expect that the rated SPL is based on the highest loudness that doesn't distory, but if it's a semi-graceful degredation then perhaps that sort of criteria can't be perfectly established.

So I brought up the question of known circumstances of inverted polarity on SteveHoffman.tv. Man, those guys are cagier than Head-Fi. Anyways, the conclusions I think I can make from that discussion are rather surprising.

• Nobody can provide concrete proof of endemic polarity reversal that is either detectable on the CD or verified by the producer.
• Moreover, the common consensus of two mastering engineers (Steve Hoffman and Clark Johnson) is that the polarity is pretty much impossible to determine objectively.
• The only way that polarity should be determined, in their eyes, is subjectively, by determining which polarity sounds better.
• Some OK anecdotal evidence for endemic polarity reversal: up until recently (if ever), microphones never had a standard XLR wiring scheme, and the hot wire could be in one of several places depending on the mic.
• In another thread it was noted that the MFSL remaster of The Mothers of Invention's "We're Only In It For The Money" was inverted with respect to the earlier Rykodisc remaster.

Don't you mean positive displacement on the basilar membrane, after going through the "cochlear filter", rather?

In any case, I wonder if it's possible to instrument the headphones to find out just how well the reproduction and inverted reproduction track each other. I've not found many devices (some few headphones) that are linear enough to attempt a test like this. Most, especially on things that are abrubt in time, tend to respond differently to positive and negative excitation.

To that, add the fact that the air itself is quite nonlinear if the peaks get to any substantial level, and I'd be worried that the air, intself, is changing the stimulus at the eardrum. Once that happens, differences aren't very surprising.

Sorry to be so late to the table here, but life is busy.

I must also note that if there are multiple microphones in the same soundfield, inverting one microphone will create substantially different filtering than it otherwise would.

I seem to recall the people doing "Perceptual Soundfield Reconstruction" stating that they had to be very sure that all of the mikes were wired and responding properly in terms of polarity, as well. That is, perhaps, less surprising.

That has to be one of the most funny thing I've read on these forums

on polarity reversal in recordings: it does happen, and fairly often. don't know about CDs, but try ripping a DVD and comparing the audio tracks.

obviously we can expect 5.1 mixes to be a fair bit different to 2.0 mixes, as they are done at different times, and possibly by different people on different equipment. however, absolute phase inversion is very often visible in places it shouldn't ought to be (centre channel of a 5.1 compared to 50/50 average of left and right in the 2.0 track). i see this nearly every day.

also, dubbed and music/effects tracks are often in opposite phase, though different language dubs tend to be in the same phase (like english + japanese). this can be spotted when playing a tape with 4 channels - full mix and music+effects. on a DVW-A500 deck, there's a short crossfade when monitoring channels are switched. with inverted polarity, this normally unnoticable crossfade becomes a dip in volume as the channels cancel out. i've observed this a few times before, though i can't remember what on (how convenient ) mainly because i didn't care and wasn't looking for it. this is obviously only relevant with newer material where the dub and M&E channels are in sample-accurate sync.

i'm not sure if the effect is present on this title, but it's the only one i can think of that has both M&E and dialogue mixes on it - the region 4 release of "Mars Attacks"

it's a real mixed bucket with video material.

A Speculation Regarding Perception of Detail

I’ve observed that if an audio system sounds good, no single component of that system can be all that bad nor can the polarity of the recording be played inverted from the live performance.

I have come to this conclusion because I haven’t been able to compensate for a bad component without causing some egregious sonic and musical tradeoffs. On the other hand, if a system really sounds awful it may only be a single component or the inverted polarity of the recording that’s causing the problem. For example, simple as it may seem, a single component could degrade the sound if its power cord is plugged into the wall outlet in less than the best sounding orientation.

A great sounding system is the result of its creator’s choice of components and musical judgment. The only true basis for their judgment is an understanding of music and a memory of unamplified acoustic instruments and voices in a reasonable acoustic venue and heard from an aesthetically correct distance.

I believe that every choice one makes in the design of an audio system involves tradeoffs, and the only question is which tradeoffs each of us finds acceptable. Around fifteen years ago, when I first became interested in the audibility and importance of absolute polarity, the speaker system that I’d created some ten years earlier and used for all my serious testing and musical enjoyment had second-order 12 dB Linkwitz-Riley crossovers. Despite its many advantages it also had one major disadvantage; it wasn’t phase coherent. Without phase coherence it was impossible for me to discern polarity or to hear music purely in or out of absolute polarity because that crossover requires some of its drivers to always play in opposite relative polarities to each other. As a result that speaker system was inconsistent with the single absolute polarity of live music. I listened to each separate driver connected first in one polarity and then the other. It wasn’t all that easy in the beginning to hear the differences, especially with my sealed back electrostatic tweeters. But since they crossed over at a relatively low 1.6 kHz I eventually decided that they, as well as all the other drivers, sounded better connected in absolute polarity. And next, with all the drivers playing in absolute polarity, I determined that I greatly preferred hearing music in absolute polarity. And from that day to this, I only find music played in absolute polarity to be truly emotionally satisfying and believe that the single most important sonic and musical aspect of a properly connected audio system is its ability to reproduce the polarity of live music.

Audio systems must at the very least satisfy the following three requirements to be suitable for rendering polarity judgments. 1. The playback polarity of the source is heard in the same polarity as the original recorded source. 2. The system is phase coherent and preferably minimum phase. In the analog domain the only classic crossover networks that permit a speaker to preserve the phase-polarity of the input signal are 6 dB first-order Butterworth. If you’re not sure about your speaker system, you may use single driver headphones. 3. The system’s frequency response deviates no more than +/- 3 dB from flat between 50 and 8 kHz which is an example of an application of the rule of 400 as defined in the first edition of the Audio Cyclopedia.

The gist of my speculation regarding the perception of detail and polarity is as follows: When one watches film, video or computer monitors the pictures are not seen as a series of separate still images and thank goodness! It’s because the frames change or refresh fast enough, typically 24, 30 or 60 plus times per second respectively. The actual flash rate may be up to120 frames per second, depending upon the medium, which causes our eye-brain’s persistence of vision to merge one still frame of a picture into the next.

Similarly, in audio, active noise-canceling headphones illustrate the ear-brain’s persistence of hearing with regard to high frequencies. The way that active noise-canceling headphones work is by picking up ambient noise with built-in microphones and then generating a signal that’s exactly out of phase to the ambient noise that, at least in principle, should cancel it completely. The specifications of active noise-canceling headphones indicate that they cancel bass frequencies much more effectively than high frequencies. And perhaps that’s true to some extent, but much of the reduction in their apparent effectiveness at higher frequencies may be the result of our ear-brain’s persistence of hearing that merges the rapidly alternating relative phase of high frequencies into one sound that has no apparent phase/polarity. Thus two tweeters playing high frequencies out of relative phase aren’t heard as canceling each other. Perhaps, this phenomena could be considered a corollary of the Fletcher-Munson effect whose curves describe the reduction of our ear’s sensitivity at both frequency extremes. Well wouldn’t it be great if noise-canceling headphones canceled the high frequencies as well as the bass frequencies? I’d surely like that and I bet you would too. So from Shakespeare – Julius Caesar, Cassius speaking, "The fault, dear Brutus, is not in our stars [equipment], but in ourselves…"

In my experience it’s exceptionally difficult if not impossible to determine solely by ear the phase of a tweeter’s electrical connection because the phase of the high frequency signal reverses too rapidly for our ear-brain to get a fix on it. In other words, when the highpass frequency of a tweeter is raised it eventually becomes so high that it exceeds our ear-brain’s ability to distinguish the phase of its electrical connection, and its rapid phase reversals merges both phases into a single sonic impression that’s without a discernable phase. For example, when two tweeters are playing a 10 kHz signal, if your head is a mere 3/10ths of an inch (a quarter wavelength) closer to one tweeter than the other, the signal from one tweeter arrives at your ears 180 degrees out of phase with the signal from the other tweeter. Although theoretically they should cancel each other perfectly, I believe most listeners will still hear the 10 kHz signal at full volume.

Before I state the conclusion of my theory you need to know something about the use of test equipment to determine polarity. The measurements of spectral content, frequency balance, dynamic range, distortion made on components playing music are the same regardless of the polarity of its playback. Were it otherwise, I wouldn’t have written this think piece about how music played out of absolute polarity affects our perception of detail. According to "The Wood Effect" many listeners can detect the polarity of asymmetrical musical signals even though test equipment and computer programs can’t. Therefore it shouldn’t seem so contrary to common sense, scientific analysis or the least bit mysterious that measurements frequently don’t correlate well with subjective listening tests. But on the other hand, perhaps some measurements will be more relevant to the way we hear when equipment is played in absolute polarity!

Now it follows, although the cutoff threshold may vary among individual listeners, as the sound’s frequency increases, above some point all listeners will perceive the music’s high frequencies as equally loud regardless of their actual polarity. But when music is heard out of absolute polarity, the midrange, bass, and even the high frequencies below some frequency, all tend to sound somewhat recessed, rather dry, and bleached out. Thus psycho-acoustically against a background of a sucked out and a papery dry sounding midrange, and a sucked out dry sounding bass, the high frequencies are heard in bold relief and sound a bit harsh, which also makes the bass and mid-range seem more detailed with faster transients, although they are not. And that can make the bass sound as if its attacks are quicker because what’s heard as the leading edge of its transients are really the sound of its harmonics which are actually reproduced by the mid-range and tweeter not the woofer. The result gives the impression of a greatly degraded stereo image that’s rather two-dimensional with a soundstage that’s vaguely focused and somewhat confused. But those effects are really only psychoacoustic artifacts of the music being played out of absolute polarity and not how acoustic instruments and voices sound live!

Here are some other examples of how the psychoacoustics of audio affects our perceptions, which sometimes seems counter intuitive, but nevertheless may resonate with some listeners. I believe when you add a subwoofer to a system it doesn’t necessarily sound as if you’ve added more bass, but more often than not, it sounds as if the highs have been reduced. Similarly, add a super tweeter to a system and it may sound as if there’s less bass not more highs. And if you turn off the bass/mid-bass altogether or reduce your mid-range, the sound seems more detailed when it’s not.

Music played out of phase coherence or out of absolute polarity may cause some listeners to wrongly attribute the low fidelity unpleasant sound to solid state devices or the digital process in general. This causes some listeners to prefer what they think is the more tuneful, full bodied, and rounded sound that they associate with tube equipment or vinyl records which they believe sounds more like a live performance, when in point of fact, all they really need to hear is music played in absolute polarity. High-fidelity equipment, tube or solid state, shouldn’t impose a sonic character of its own on the musical signal; its only tasks are to amplify the signal without distortion and control the speakers! How much tweaking and component swapping in our systems are only musically misinformed attempts to correct for music played out of absolute polarity that in Absolute Reality are bound to fail the test of high-fidelity? Does this suggest that the conclusions of some prior listening tests should be reevaluated and repeated with music that we know for sure is played in absolute polarity? I definitely think so, and that should include recordings as well, but each of you may answer that question for yourselves.


The Louis Objective Test of the Audibility of Relative Polarity

Someone, other than the test subject, compiles a 72-minute test CD-R or CD of 72 one-minute music tracks as follows: The first track will be a one-minute excerpt from a CD, record or tape recording of a two microphone stereo recording selected for its musical value but without regard to its actual polarity. The second track will be the same one-minute excerpt as the first track, but it will be recorded in the same polarity as the first track or in the opposite relative polarity to the first track as determined by the toss of a coin, heads the same and tails in the opposite polarity. The same procedure is followed for tracks 3 and 4 with different music and repeated again until 36 pairs of identical music tracks have been recorded to the CD for a total of 72-minutes. The person making the disc memorializes the polarity of each of the even numbered tracks relative to its odd numbered counterpart, and thus he has created a test CD for The Louis Objective Test of the Audibility of Relative Polarity. The playback system should use single driver headphones or at least a speaker system with consistent polarity, i.e. all drivers move in the same relative phase to each other. The actual polarity of any track or the playback system doesn’t affect the validity of the test because it’s only a test of the audibility of relative polarity not absolute polarity. A test of a person’s ability to discern the actual polarity isn’t necessary if they can’t pass the relative polarity test. However, test subjects who prove they discern absolute polarity would obviously pass a relative polarity test.

There’s also The Louis Objective Test of the Audibility of Absolute Polarity, but unlike the test for relative polarity one must know the polarity of the recordings that comprise the compilation test disc. I know how to establish the polarity of recordings, but since I’m not ready just yet to explain how that’s accomplished the more difficult test for the ability to discern actual polarity will have to wait just a little while longer.

The standard way to scientifically compare component A against component B is by double blind ABX testing. In order to make ABX testing a bit easier I’ve added non-X that allows the test subject to hear non-X but only knowing that it’s non-X. When the purpose of the test is to find whether component A or B is preferred, I have a more direct protocol. Component A and component B are played alternately double blind first one and then the other as many times as needed to state a preference. Then the sequence is repeated with the order of A and B chosen at random for the next set of alternating comparisons. The protocol is repeated until the results are statistically significant. This single test will determine directly both the test subject’s ability to distinguish A from B as well as preference.

Best regards,

George, Perfect Polarity Pundit™ Copyright 2006 All rights reserved by George S. Louis

Digital Systems & Solutions
Hear the Music Not Just Notes™
1573 Kimberly Woods Drive, El Cajon, California 92020-7261
Telephones: 619-401-9876, 888-588-9542
Fax: 619-588-5524, e-mail: gslouis@gslouis.com
Word of Ear Products from the Company that’s Sound Millenniums Ahead™

Can you please explain how the performance of AC power over a metallic conductor can vary with conductor direction. This claim would require some impressive proof.

The crossovers require you to connect some of the drivers in reverse polarity because they, themselves, reverse the phase. By connecting the drivers reversed, you restore the phase. If you didn't connect them the wrong way round, you would get a very big dip (sometimes -40dB or more) in the response in the crossover region. Crossover designers refer to this as a "reverse null" and it's depth and curves can tell you a lot about the crossover design.

+-3dB is a large amount of ripple in the pass band of your system. All the active components should be able to achieve +- 0.1dB or less over this range. Speakers are harder, but a lot of designs still do much better than this on-axis. 6dB peak-to-peak ripple in the response will likely be clearly audible as a coloration of the music or change of timbre.

Your three criteria for good sound miss out a lot of important (and audible) parameters in system design. Noise is one that you don't mention, as is total harmonic distortion (THD), intermodulation distortion (IMD) and a variety of other measurements. I think you are over-emphasising the audibility of linear effects compared to that non non-linear effects.

If the two were point sources, and you put them an integer number of wavelengths apart, then there would be places where the sound is cancelled and places where it is re-inforced. With real tweeters the picture is much, much more complex, but you will still get areas where there are peaks and areas where there are nulls.

Computer programs certainly can, but it requires you to know a little about the signal you are dealing with. Vocal recordings, for example, often have a very assymetrical look to them, that it would be easy to identify with a computer.

Your test procedure is interesting, and it would be nice to see it implemented.

as i was eletronic technician i can write:
wonderful post georgelouis,a long time i don't read clever explanations.
from the start i could feel that you have experience.[/quote]For example, simple as it may seem, a single component could degrade the sound if its power cord is plugged into the wall outlet in less than the best sounding orientation. [/quote]only experienced persons knows this details.

i don't want to quote parts from your post but some are very interesting because the users always forget /ignore important details when are doing tests(in ABX tests sometimes too).
in general i don't use statistic ABX tests as "final reference with scientific proove" because who apply the test,ignore or never knew details like the "three requirements for audio system" explained in your post to tell the minimum.
in general,how can we be sure of the "consensual scientific proove" with ignored details outside the test?
nothing against the tests but they are incomplete!

i'm sad because i can't write in plain english like you but i can read and understand.
have more people that knows what he only read somewhere that people that knows what he learn and have experience.(here you can read my crap english)

welcome and best regards.

ABX procedures or tools are used by people at home to test audibility of differences with their own equipment, an in this context they are perfectly valid. They are "personal" tests. Often, people with very good equipment don't get very different results.

Professional researchers, on the contrary, use reference level equipment for their double-blind tests, be in the form of ABX, ABC/HR or other variant.

i agree with you, you wrote clever but the same as me.
(now i'm trying to explain)...
i wrote that "in general i don't use statistic ABX tests as "final reference with scientific proove" because who apply the test,ignore or never knew details like ...
means:
"in general" =don't means that i don't use.
"final reference with scientific proove" = general consensus. some people(lots here) knows how to use,others are sure that knows but commit mistakes.
"statistic" and "who apply the test" = results from group of persons doing tests but ignore or...etc.
(i told,crap english here)

ABX,ABC/HR are perfects only for who knows what is doing...
no one knows ABX better than you and i'm right that you know what i mean now.

...and forgive my english!
regards.

Finally! George found my thread. I was wondering when you'd get around to it.

I don't have a whole lot of time right now to reply to all of that (I ought to have been in bed an hour ago). But I will say that, if I had to do the polarity test again, except with 1-minute tracks to compare blindly, I believe I would have absolutely failed. I had to change tracks much, much faster than that to get any sort of consistency. I don't think I could have done it without computer-aided ABX testing, and foobar in particular. (Then again, I don't consider myself a particularly detailed listener. I fail ABXing of 128kbps encodes on a regular basis.)

I usually zeroed in on a roughly 5-second section of the music and listened to each version repeatedly, as fast as possible, but occasionally I listened to a selection roughly half a second long and alternated between the two half-second tracks as fast as possible.

The scientific literature I'm aware of (but alas can't quote off the top of my head) says that audio memory is strongest in the last few seconds, and you really start to forget things after a minute. They're not kidding. That completely supports my experience.